Biomedical
Signal
Processing
and
Control
8 (2013) 59–
65
Contents
lists
available
at
SciVerse
ScienceDirect
Biomedical
Signal
Processing
and
Control
journa
l
h
omepage:
www.elsevier.com/locate/bspc
ECG
compression
using
the
context
modeling
arithmetic
coding
with
dynamic
learning
vector–scalar
quantization
Boqiang
Huang
a,b
,
Yuanyuan
Wang
a,∗
,
Jianhua
Chen
c
a
Department
of
Electronic
Engineering,
Fudan
University,
Shanghai
200433,
China
b
Institute
of
Mathematics,
Paderborn
University,
Paderborn
33098,
Germany
c
Department
of
Electronic
Engineering,
Yunnan
University,
Kunming,
Yunnan
650091,
China
a
r
t
i
c
l
e
i
n
f
o
Article
history:
Received
2
November
2011
Received
in
revised
form
11
February
2012
Accepted
3
April
2012
Available online 17 May 2012
Keywords:
ECG
compression
Vector
quantization
Scalar
quantization
Context
model
Conditional
entropy
coding
a
b
s
t
r
a
c
t
Electrocardiogram
(ECG)
compression
can
significantly
reduce
the
storage
and
transmission
burden
for
the
long-term
recording
system
and
telemedicine
applications.
In
this
paper,
an
improved
wavelet-based
compression
method
is
proposed.
A
discrete
wavelet
transform
(DWT)
is
firstly
applied
to
the
mean
removed
ECG
signal.
DWT
coefficients
in
a
hierarchical
tree
order
are
taken
as
the
component
of
a
vector
named
tree
vector
(TV).
Then,
the
TV
is
quantized
with
a
vector–scalar
quantizer
(VSQ),
which
is
composed
of
a
dynamic
learning
vector
quantizer
and
a
uniform
scalar
dead-zone
quantizer.
The
context
modeling
arithmetic
coding
is
finally
employed
to
encode
those
quantized
coefficients
from
the
VSQ.
All
tested
records
are
selected
from
the
Massachusetts
Institute
of
Technology-Beth
Israel
Hospital
arrhythmia
database.
Statistical
results
show
that
the
compression
performance
of
the
proposed
method
outperforms
several
published
compression
algorithms.
© 2012 Elsevier Ltd. All rights reserved.
1.
Introduction
Since
the
electrocardiogram
(ECG)
has
great
clinical
significance
in
diagnosing
heart
diseases,
it
is
extensively
used
in
many
situa-
tions
such
as
the
24-h
portable
Holter
monitor,
the
clinical
ECG
workstation
and
telemedicine
applications.
However,
the
amount
of
digital
ECG
data
grows
with
the
increase
of
the
sampling
rate,
the
sample
resolution,
the
recording
time,
the
number
of
chan-
nels,
the
number
of
subjects,
etc.
It
gradually
becomes
a
problem
in
these
applications
when
the
storage
space
and
bandwidth
are
very
limited.
Therefore,
an
effective
data
compression
method
attracts
much
attention
from
many
researchers.
On
the
other
hand,
charac-
teristics
of
the
ECG
signal,
including
the
heart
rate,
the
PR
interval,
the
QRS
duration,
the
QT
interval,
etc.,
are
the
important
evidence
for
doctors
to
diagnose
diseases.
Any
obvious
variation
of
the
ECG
waveform
induced
by
ECG
processing
algorithms
may
cause
a
mis-
diagnosis.
Thus,
the
target
of
ECG
compression
is
to
reduce
the
redundancy
as
much
as
possible
while
to
maintain
clinically
accept-
able
signal
quality.
During
last
three
decades,
lots
of
ECG
compression
methods
have
been
presented.
Generally,
they
can
be
divided
into
three
∗
Corresponding
author.
Tel.:
+86
21
65642756;
fax:
+86
21
65643526.
E-mail
addresses:
bhuang@math.uni-paderborn.de
(B.
Huang),
yywang@fudan.edu.cn,
yywang@fudan.ac.cn
(Y.
Wang),
chenjh@ynu.edu.cn
(J.
Chen).
categories:
direct
methods
[1–3],
parameter
methods
[4,5]
and
transform
methods
[6–13].
A
detailed
summary
of
these
meth-
ods
is
described
in
[8].
Since
the
wavelet
transform
has
the
good
time–frequency
localization,
the
wavelet-based
compression
algo-
rithm
achieves
the
great
performance
among
transform
methods.
There
are
three
representative
wavelet-based
lossy
compression
algorithms
for
ECG
signals.
The
first
one
is
the
compression
based
on
the
set
partitioning
in
hierarchical
trees
(SPIHT)
[6].
It
modi-
fies
the
premier
state-of-the-art
algorithm
in
image
compression
to
suit
special
characteristics
of
ECG
signals.
Secondly,
Miaou
et
al.
proposed
a
SPIHT
based
compression
method
using
the
vector
quantizer
(VQ)
with
a
distortion
constrained
codebook
replenish-
ment
[11],
in
which
the
codebook
of
the
VQ
can
update
itself
to
match
the
unknown
input
source.
Since
the
VQ
is
a
mapping
from
a
vector
in
a
k-dimensional
Euclidean
space
into
an
integer
called
the
codeword,
the
modified
method
outperforms
the
original
SPIHT
based
method.
The
third
method
is
the
one
using
context
modeling
arithmetic
coding
(CMAC)
with
a
uniform
scalar
dead-zone
quan-
tizer
(USDZQ)
[13].
It
extracts
precise
conditional
information
in
the
coded
neighborhood
for
each
symbol
to
be
coded,
which
obvi-
ously
improves
the
efficiency
of
the
entropy
coding.
Although
the
third
method
employs
the
scalar
quantizer,
statistical
experimen-
tal
results
show
that
its
compression
performance
is
close
to
that
of
second
one.
To
modify
existing
ECG
compression
algorithms,
the
proposed
method
aims
to
take
advantages
of
both
VQ
and
CMAC.
To
main-
tain
the
vector
form
in
[11],
the
tree
vector
(TV)
is
extracted
from
1746-8094/$
–
see
front
matter ©
2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.bspc.2012.04.003